Fragile States and

Development Policy∗

Timothy Besley London School of Economics and CIFAR

Torsten Persson

IIES, Stockholm University and CIFAR January 2011

The Suntory Centre Suntory and Toyota International Centres for Economics and Related Disciplines London School of Economics and Political Science Houghton Street London WC2A 2AE

EOPP/2011/22 Tel: (020) 7955 6674 ∗ This paper is based on Besley’s Presidential Address to the European Economic Association presented at the 25th Congress of the Association held in Glasgow in August 2010. We thank Paul Collier, Fabrizio Zillibotti and an anonymous referee for comments on an earlier draft. We are grateful to CIFAR, the ESRC, the Swedish Research Council, the Torsten and Ragnar Söderberg Foundation, and the ERC for financial support.

Abstract

It is widely recognized that fragile states are key symptoms of under-development in many parts of the world. Such states are incapable of delivering basic services to their citizens and political violence is commonplace. As of yet, mainstream development economics has not dealt in any systematic way with such concerns and the implications for development assistance. This paper puts forward a frame-work for analyzing fragile states and applies it to a variety of development policies in different types of states. JEL Codes: 010, 019 and P45 Keywords: state fragility, development

This series is published by the Economic Organisation and Public Policy Programme (EOPP) located within the Suntory and Toyota International Centres for Economics and Related Disciplines (STICERD) at the London School of Economics and Political Science. This new series is an amalgamation of the Development Economics Discussion Papers and the Political Economy and Public Policy Discussion Papers. The programme was established in October 1998 as a successor to the Development Economics Research Programme. The work of the programme is mainly in the fields of development economics, public economics and political economy. It is directed by Maitreesh Ghatak. Oriana Bandiera, Robin Burgess, and Andrea Prat serve as co-directors, and associated faculty consist of Timothy Besley, Jean-Paul Faguet, Henrik Kleven, Valentino Larcinese, Gerard Padro i Miquel, Torsten Persson, Nicholas Stern, and Daniel M. Sturm. Further details about the programme and its work can be viewed on our web site at http://sticerd.lse.ac.uk/research/eopp. Our Discussion Paper series is available to download at: http://sticerd.lse.ac.uk/_new/publications/series.asp?prog=EOPP For any other information relating to this series please contact Leila Alberici on: Telephone: UK+20 7955 6674 Fax: UK+20 7955 6951 Email: l.alberici @lse.ac.uk © The authors. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source.

1 Introduction

The donor community has increasingly begun to confront the implicationsfor development policy of the prevalence of weak, fragile, and failing statesin some of the world�s poorest countries. And this has lead to a gradual shiftin focus towards the speci�c problems that arise in such states (see, e.g.,McGillivray, 2006). As a result, there is now an emerging policy literaturewhich worries about how to deal with such states and what can be done toimprove their situation (see, e.g., OECD, 2010a).Speci�c lists of fragile states have been produced by national aid agencies,

international organizations, and individual research teams.1 To illustrate,consider the 2009 Polity IV classi�cation of fragile states which is summarizedin Figure 1.2 In this �gure, countries are classi�ed according to their decile inthe distribution of fragility. Countries with the greatest fragility are coloredin black while those with least fragility are in white. Intermediate deciles arecolored in shades of gray. According to this classi�cation, the most fragilestates are found in Subsaharan Africa and South Asia.The Polity IV project on which Figure 1 is based is admirably clear in

explaining its eight underlying criteria, which include factors such as securitye¤ectiveness, political legitimacy, and economic e¤ectiveness. These conceptsare all based on speci�c empirical indicators that are aggregated in a trans-parent way. Such indicators certainly create a basis for debate about whichfactors shape state fragility. But the conceptual underpinning of the mea-surement is far from clear, and sometimes appears to confuse symptoms andcauses. For example, low income is included as a component of state fragility.But is poverty a product of state fragility or a cause? The same can be saidfor the measures of violence or insecurity, which also �gure prominently inthe de�nition.Typically, development agencies are less clear about their basis for clas-

sifying a state as fragile. For example, USAID (2005) uses two criteria, eachof which is su¢ cient to have a state classi�ed as fragile: vulnerability andcrisis. Vulnerability refers to those �states unable or unwilling to adequatelyassure the provision of security and basic services to signi�cant portions oftheir populations and where the legitimacy of the government is in question.

1Looking at di¤erent classi�cations of fragility (however termed) leads to rather similarrankings of states.

2Rice and Patrick (2008) give an overview of di¤erent attempts to measure state weak-ness or fragility.

2

This includes states that are failing or recovering from crisis.�Crisis refersto �states where the central government does not exert e¤ective control overits own territory or is unable or unwilling to assure the provision of vital ser-vices to signi�cant parts of its territory, where legitimacy of the governmentis weak or nonexistent, and where violent con�ict is a reality or a great risk.�As far as we can tell, these criteria are applied subjectively.Notwithstanding the attention to fragile states by the policy commu-

nity, mainstream development economists have paid little attention to theseproblems in spite of a resurgence of interest in the political economics ofdevelopment.Our paper tries to take some steps in this direction by exploring the ori-

gins of state fragility and highlighting two main pathologies that need to beexplained: state ine¤ectiveness in enforcing contracts, protecting property,providing public goods and raising revenues, and political violence either inthe form of repression or civil con�ict. We propose a theoretical framework,in which the roots of these pathologies can be explored.3 In this framework,a government is endowed with a level of �scal capacity (the ability to tax)and a level of legal capacity (the ability to support markets). Policy decisionsinclude how much to spend on public goods and transfers. Political institu-tions may constrain the distribution of transfers; stronger, more cohesive,institutions prevent incumbents from using the state to make transfers solelyto the incumbent group. Governments can make three kinds of investments:(i) in �scal capacity; (ii) in legal capacity, and (iii) in violence, the latter as ameans of holding on to power. Opposition groups can also invest in violenceif they choose to, and do so in order to acquire power.Seen through the lens of our model, an ine¤ective state is one which

has made few investments in �scal and legal capacity, and a violent state isone where the government and opposition invest in violence to maintain oracquire political power. A common interest in providing public goods, fos-tered either by circumstances or cohesive political institutions, can eliminateboth of these problems. However, when institutions are non-cohesive eitherpathology may emerge and the model allows us to explore the conditions fortheir emergence, thus uncovering the roots of state fragility. Having an ex-plicit theory enables us to clarify what is exogenous and what is endogenous.

3The paper builds on our earlier work, especially Besley and Persson (2009) and (2010).Besley and Persson (2011) provide a full exploration of the ideas developed here, and adiscussion of microoeconomic and micropolitical foundations for the reduced-form relationsused in this paper.

3

It also enables us to clarify what is a symptom and what is a cause �a verymurky distinction in the present policy debate.The main output of our theoretical model is a matrix of parameter ranges,

de�ning the roots of di¤erent state pathologies. States that are peaceful andhave high levels of state capacity have common interests and/or cohesivepolitical institutions. In the absence of such commonalities, however, thedetails really matter. Pathologies are identi�ed with cases in which there iseither repression or civil war on the one hand, and a weak or redistributivestate on the other. On the basis of this observation, we suggest an AnnaKarenina Principle of state organization, paraphrasing the 1st line of LeoTolstoy�s 1870s novel):

�All happy families resemble each other; every unhappy family isunhappy in its own way.�

In our case, common-interest states are the happy families with peace as wellas high state e¤ectiveness. As in the Tolstoy quote, the unhappy familiescome in many forms. This principle re�ects a real, practical di¢ culty indealing with fragile states, namely to understand the factors that are drivingthe state in a particular direction.Our theoretical framework helps to sharpen the policy discussion by high-

lighting how di¤erent factors contribute to state fragility. The focus on frag-ile states underlines a wider set of issues, which are crucial to discussionsabout development policy. It is well understood that the problem of under-development re�ects a complex array of interdependent factors which clus-ter together, and where low income and living standards make up just onedimension. The notion of a fragile state is useful in highlighting this multi-dimensionality, and in getting away from excessive emphasis on income andinsu¢ cient emphasis on state ine¤ectiveness and political violence.Our analysis also serves to highlight some of the policy dilemmas as it

allows us to think about the consequences of development assistance. Specif-ically, we use our model to analyze the e¤ects of various forms of assistance invarious types of countries. This analysis takes into account the equilibriumresponses in the recipient country in terms of policymaking, state-capacityinvestments and investments in violence and how these depend on the formof intervention as well as any speci�c pathologies with regard to state inef-fectiveness or the propensity for political violence of di¤erent types.A large amount of aid is given by rich countries to nations in the devel-

oping world. O¢ cial Development Assistance (ODA) comes mainly from the

4

23 members of the Development Assistance Committee (DAC). Accordingto the OECD, the total �gure in 2009 is around $123 billion (this �gure,as others below, is in �xed 2008 prices), which is the highest number everrecorded. While the amount of aid kept increasing in the post-war period, itstarted falling immediately after the end of the cold war, but then picked upagain. In 2002 it surpassed its previous peak of $86 billion from 1992, andhas kept increasing since then. While aid targets for rich countries have beenset at 0.7% of Gross National Income, very few countries meet these targetsand, in fact, the trend is declining over time. The amount of aid from theDAC countries today stands at about 0.31 of their total GDP, as opposed to0.54 in 1961.Out of the $120 billion of total ODA given in 2008, about 30% was given

indirectly through contributions to multilateral institutions like the Euro-pean Union and the World Bank. About one sixth came in the form ofTechnical Cooperation (plus additional indirect amounts through multilat-eral institutions). On top of ODA, additional aid is given through the privateand institutional sector, with more than $20 billion channeled by NGOs. Thelargest regional recipient of aid is Sub-Saharan Africa, which received 33% ofall ODA in 2007-2008, followed by 21% to Middle East-North Africa, about15% each to South-Central Asia and the rest of Asia, 9% to Latin America,and 4% to Europe.4

Exactly what can be done to improve the well-being of citizens in poornations remains a controversial topic.5 A shining example, which buoyedinterest and enthusiasm for aid, was the experience of the Marshall planfor rebuilding post World War II Germany and other parts of Europe. Be-tween 1948 and 1951, the U.S. transferred around $13 billion to Europeaneconomies. This episode created a sense that large-scale resource transferscould make a signi�cant di¤erence to economic development, a sense thatwas further underpinned by the so-called Truman Doctrine, which called fora global focus on the plight of the developing world.Arguably, the Marshall plan fuelled the belief that lack of resources is the

key impediment to economic development and aid �ows are necessary to buildpublic institutions and stocks of capital. Countries may eventually achievea successful development path if left to themselves, but a helping hand of

4See OECD (2010b) for these background �gures. Gupta, Patillo, and Wagh (2006),give an overview of trends in overall aid �ows during 1960-2004.

5See Riddell (2007) for a review. Temple (2010) provides an excellent review of manyof the economic debates.

5

international transfers would speed up that progress. Chenery and Strout(1966) is a key exposition of the underlying ideas, and a modern statementof a similar view is Sachs (2005).The real-world experience has not ful�lled the rather romantic vision of

early aid traditionalists. Aid pessimists point to the fact that much of aidwould not survive any reasonable cost-bene�t test. Domestic political agendasof governments in poor countries have frequently not supported economicdevelopment, and these governments often lack the technical competenceto spend resources wisely. Because if this, it is argued, much aid is wastedand does not contribute to developmental ends. Bauer (1972, 1975) was anearly aid pessimist, while a modern aid critique is strongly stated in Easterly(2006).6

More recently, some observers have attempted to reconcile the two views,by being more optimistic on conditionality and the ability of analysts to iden-tify the underlying pathologies. Certainly, traditionalists have had a naiveview of the workings of political institutions or the ability to navigate aroundpolitical constraints. Without a more forthright analysis of the institutionalenvironment, it is hard to make progress. Collier (2007) can be seen as anexponent of this revisionist view.Our attempt in this paper to map out di¤erent kinds of states and the

consequences of di¤erent forms of development assistance can be seen asan e¤ort to elucidate such a revisionist view. Even though our explorationis stylized and theoretical, the model gives a clear sense of the complexityof the issues and policy dilemmas. The exact impact of aid will likely varywith institutions as well as circumstances. Generalizations about the bestcourse of action are impossible given the heterogeneity in experience and thelimited empirical knowledge about several of the channels uncovered by ourtheoretical analysis. This underlines the di¢ culties the donor communityhas to grapple with when facing the problems of fragile states.The remainder of the paper is organized as follows. In Section 2, we

lay out a canonical two-period model with investments in state capacityand political violence. Section 3 puts the model to work by investigatingthe e¤ects of di¤erent forms of development policy �beginning with cashtransfers, but branching out to other forms like technical assistance and post-con�ict assistance. In Section 4, we sketch some extensions of the analysis.

6Djankov, Montalvo and Reynal-Querol (2008) argue that aid leads to a deteriorationin the quality of institutions.

6

Section 5 concludes.

2 Theoretical Framework

The modeling approach draws on our earlier work: Besley and Persson (2009,2010, 2011).7 We develop a two-period approach to investments in statecapacity and violence.

2.1 Basic set-up

There are two time periods s = 1; 2 with two groups of individuals, A and B;each of which has comprises half the population. Every individual has wagerate !(�s) where ! (�) is an increasing, concave function, and where �s is thegovernment�s legal capacity. There are no savings.At the beginning of s = 1; one group holds power and we will refer to

this group as the incumbent I1 2 fA;Bg. The other group is the oppositionO1 2 fA;Bg: Between the two periods, there can be a transition of power.This is a¤ected by investments in political violence by the incumbent andopposition. In period 1; the opposition group O1 can mount an insurgencywith army LO � LO; paid for within the group, at marginal cost of funds �.We interpret � as a reduced form representation of how well-organized andwell-funded is the opposition. The incumbent group I1 can invest in armyLI � L

I; paid out of the public purse, at marginal cost �1. There is no

conscription: each soldier is just paid the period-1 wage, namely !(�1).The probability that the opposition takes over depends on the con�ict

technology (LO; LI ; �) which is increasing in LO; and decreasing in LI . Thewinner becomes next period�s incumbent, I2 2 fA;Bg and the loser becomesthe new opposition, O2 2 fA;Bg. If nobody takes up arms, then the tran-sition probability is (0; 0; �). The parameter � is an index of the marginalgain from �ghting �see Assumption 2 below.We assume that utility is linear and depends on the quantities of public

and private goods:uJs = x

Js + �sgs ,

7Prior contributions related to the analysis in this paper include Acemoglu (2005),Bates (2008), Caselli (2006), and Weingast (2005).

7

where cJs denotes private consumption of a group-J member at s; and gs is theconsumption of public goods with �s denoting their value. One archetypalexample of gs is "defense", in which case �s is the "threat of external con�ict".Private consumption in period s is

xJs =

�(1� � 1)! (�1) + rJ1 � �J�! (�1)LO at s = 1(1� � 2)! (�2) + rJ2 at s = 2 .

where rJs is a transfer targeted towards group J and �J = 1; if J = O; and

0 if J = I: The variable � s is �scal capacity, a costly investment made oneperiod ahead.8

The value of public goods stochastic. We assume a two-point distribution�s 2 f�L; �Hg; where �H > 2 > �L > 1; and Prob[�s = �H ] = �. Shocksto �s are iid over time and we assume that the realization of �s is knownwhen policy is set. This captures uncertainty over the use state resourcesin future which depends on the demand for common-interest public goods.For example, the citizens may not know whether there will be a threat ofexternal war in the future.The period-1 incumbent can invest in both legal and �scal capacity. We

take the initial stocks f�1; � 1g as given and, for simplicity, assume that bothkinds of investment are irreversible. Fiscal and legal capacity can be aug-mented by non-negative investments f�2 � �1; � 2 � � 1g. Investment in legalcapacity takes the form of courts, judges, credit and property registries. Suchinvestments are associated with a convex cost: L(�2��1); where L�(0) = 0:In the case of �scal capacity, the investment can be thought of as developinga tax authority, its compliance structures and infrastructure to enforce an in-come tax (or impose a value added tax). We posit a convex cost F(� s�� s�1)of investing in �scal capacity with F� (0) = 0:Government decisions at date s comprise

�gs; frJs gJ=I;O;ms

; where

ms =

�F(� 2 � � 1) + L (�2 � �1) + ! (�1)LI if s = 10 if s = 2

is the cost of investing in legal capacity, �scal capacity and violence in period

8In our earlier work, we work with constraints on the government�s tax and regulatorypolicies: ts � � s and pJs � �s: Under the same assumptions as in this paper, we showin Besley and Persson (2011) that it is always optimal for the incumbent to fully exploitboth legal and �scal capacity in every period. Thus, we skip that step here.

8

s. The government budget constraint is9

R + � s! = gs +ms +rIs + r

Os

2; (1)

where R is any additional revenue source accruing only to government in theform of natural resource ownership and foreign aid. We will suppose that thelevel of such resources is �xed and known ex ante.Political institutions constrain the incumbent who must give a share �

(� 1) to the opposition group for any unit of transfers that it gives to its owngroup. It is more convenient to work with the parameter � = �

1+�2 [0; 1

2]

to represent more �cohesive�institutions. With � close to 1=2, there is equalsharing and � close to zero represents a situation with weak constraints. Ahigher value of � corresponds to greater checks and balances on executive orbetter representation of the opposition in government.The timing of decisions in the model is as follows:

1. There is an initial level of state capacity f� 1; �1g and an incumbentgroup I1:

2. Nature determines the period-1 value of public goods: �1:

3. The incumbent chooses period-1 policies frI1; rO1 ; g1g and investmentsin period-2 state capacities � 2 and �2. Simultaneously, the incumbentand opposition choose their investments in violence:

�LI ; LO

.

4. Period-1 consumption takes place.

5. The period-1 incumbent remains in power with probability 1� (LO; LI ; �)and nature determines the period-2 value of public goods: �2.

6. The period-2 incumbent chooses second-period policies frI2; rO2 ; g2g.

7. Period-2 consumption takes place.

9This formulation asssumes that �scal capacity is always fully exploited. In the un-derlying research (Beslsy and Persson 2009, 2010, 2011), we allow the government to seta tax rate on income, ts; under the constraint ts � � s: There, we show that under theassumptions of this paper, any incumbent does indeed exploit �scal capacity to the limit,ful�lling this constraint with equality.

9

We look for a sub-game perfect equilibrium in policy, violence and statecapacity investments. Since the problem has a conveniently recursive struc-ture, we are able to study these three parts separately beginning with policychoices in each period. For the violence and state capacity investment deci-sions, we work backwards with citizens forming rational expectations aboutthe period-2 policy outcome.

2.2 Policy

The optimal public spending choices�gs; r

Is ; r

Os

are made by the period s

incumbent and maximize:

�sgs + (1� � s)! (�s) + rIs ,

subject to rOs � �rIs and the government budget constraint, (1). We needto solve for two dimensions of policy: transfers and public-goods provision.To derive the optimal levels of transfers, we use the government budget

constraint to obtain:

rJs = �J [R + � s! (�s)� gs �ms] ,

where �I = 2(1� �) and �O = 2�. The revenue available for transfers is anypart not spent on public goods, gs or investments ms. Transfers are dividedbetween the two groups depending on the cohesiveness of institutions, asmeasured by �. Since �I � �O, the incumbent group will obtain a higherlevel of transfers.The optimal level of public-good provision is given by:

G (�s; � s) =

�Rs + � s! �ms if �s � 2 (1� �)0 otherwise .

All residual public revenues are spent on either transfers or public goods inperiod s; depending on the realization of the value of public goods: �s. Ifinstitutions are cohesive (� is close to one half), then all spending is on publicgoods and none is on transfers.Plugging in these optimal policies, we derive the following �indirect�pay-

o¤ function for group J in period s:

W (�s; � s; �s; Rs;ms; �J) = �sG (�s; � s) + (1� � s)! (�s) +

�J [R + � s! (�s)�G (�s; � s)�ms] .

10

For future reference, it is useful to de�ne �value functions�

U I (� 2; �2) =��W

��H ; � 2; �2; R2; 0; �

I�+ (1� �)W

��L; � 2; �2; R2; 0; �

I��

and

UO (� 2; �2) =��W

��H ; � 2; �2; R2; 0; �

O�+ (1� �)W

��L; � 2; �2; R2; 0; �

O��

for being the incumbent or opposition group in period 2 depending on thestate variables f� 2; �2g. These are the expected value of arriving in period2 with state capacities f� 2; �2g either as a member of the incumbent oropposition group.Putting these together, the expected period-2 utility of group J in period

1is:

W (�1; � 1;m1; �J) (2)

+(1� (LO; LI ; �))U I (� 2; �2) + (LO; LI ; �)UO (� 2; �2)

for the incumbent group and

W (�1; � 1;m1; �J)� �! (�1)LO (3)

+ (LO; LI ; �)U I (� 2; �2) +�1� (LO; LI ; �)

�UO (� 2; �2)

for the opposition group. For the opposition, we have deducted the cost ofviolence: �! (�1)LO; while, for the incumbent, violence is funded from thepublic purse.These payo¤s are key to understanding the determinants of investments

in violence and state capacity. We begin by study the Nash equilibriumbetween the incumbent and opposition group in their violence decisions.

2.3 Investments in political violence

The prospective trade-o¤ for the incumbent and opposition, as they contem-plate investing in violence at stage 3, is to weigh a higher chance of period-2political control against the cost of the investment.10 To study this as sim-ply as possible, and to put some structure on the problem, we impose thefollowing restrictions on the con�ict technology:

10The structure of the con�ict model used here is a natural dynamic extension of theset-up in Besley and Persson (2010).

11

Assumption 1 For all LJ 2h0; L

Ji, we have:

a. if 2 (0; 1), O > 0; I < 0, OO < 0; II > 0;b. � I(0;0;�)

O(0;0;�)� �H

�; and

c. I OO O

� IO � O II I

.

Condition a just says that �ghting always has positive returns for bothgroups, albeit at a decreasing rate. The property in b ensures that theincumbent has a higher marginal return to �ghting, when both parties donot invest in violence. It guarantees that the incumbent has a su¢ cient ad-vantage in the returns to �ghting relative to the rebels. Finally, c restrictsthe extent of any strategic complementarities or substitutabilities in the con-�ict technology. Assumption 1 is satis�ed by a number of reasonable, andcommonly used, contest functions.11

We will also make:

Assumption 2 � I��LO; LI ; �

�> 0 and O�

�LO; LI ; �

�< 0:

This says that � indexes the advantage of the incumbent: raising � increasesthe incumbent�s marginal return to �ghting while reducing the opposition�smarginal return to �ghting (in terms of each group�s probability of holdingpower in period 2):We now characterize the Nash equilibrium in violence levels denoted asn

L̂O; L̂Io. These maximize (2) and (3). The �rst-order conditions are:

I(L̂O; L̂I ; �)

�UO (� 2; �2)� U I (� 2; �2)

�� �1! (�1) � 0 ,

where �1 = max f�1; 2 (1� �)g and

O(L̂O; L̂I ; �)

�U I (� 2; �2)� UO (� 2; �2)

�� �! (�1) � 0 :

This way of writing the �rst-order conditions makes transparent that themarginal bene�t of investing in violence comes from the increased probabilityof being the incumbent in period 2, while the cost is the resources that areneeded �nance this violence whether from public or private funds. For both

11See Dixit (1987) for an overview of contest functions.

12

groups, the bene�t is proportional to�UO (� 2; �2)� U I (� 2; �2)

�, the value of

being an incumbent in period 2. The parameter �1 is the opportunity costof public funds to the incumbent. A key observation for the result to follow,is that: �

U I (� 2; �2)� UO (� 2; �2)�= ! (�1) 2 (1� 2�)Z ,

where

Z = (1� �)�R + � 2! (�2)�G (�L; � 2)

! (�1)

�is the bene�t from holding o¢ ce in terms of residual tax revenues relativeto the opportunity cost of �ghting determined by the period-1 wage. Thisvariable will determine the outcome in cases where common interests andpolitical institutions are weak.We now use the two �rst-order conditions to characterize the Nash equi-

librium and its dependence on some key parameters. Our �rst result gives aguaranteed condition for peace:

Proposition 1 If �L � 2(1 � �) or � is close to 1, neither group investsin political violence, i.e. L̂I = L̂O = 0.

This proposition says that as long as institutions are su¢ ciently cohesive, orthere is a high enough demand for public goods, there is never any politicalviolence. In this case, the marginal bene�t of being the incumbent goes tozero as there is full agreement over policy. Since investing in violence iscostly, neither group chooses to invest anything.We now explore what happens when these conditions do not hold.

Proposition 2 If Assumption 1 holds, �L < 2(1 � �) and � is below 1,there are two thresholds ZI(�; �; �) and ZO(�; �; �); where

ZI(�; �; �) = � �1 I (0; 0; �) (1� �)2(1� 2�)

< ZO(�; �; �) =�

O (0; 0; �) (1� �)2(1� 2�)

such that:

1. if Z � ZI ; there is peace with bLO = bLI = 02. if Z 2

�ZI ; ZO

�, there is repression with bLI > bLO = 0

13

3. if Z � ZO; there is civil con�ict with bLI ; bLO > 0.Moreover, bLO and bLI , whenever positive, increase in Z :

The proof is in the Appendix.The proposition describes three states of violence. When Z is below ZI , no

con�ict erupts, as both incumbent and opposition accept the (probabilistic)peaceful allocation of power, where the opposition takes over with probability (0; 0; �). When Z 2

�ZI ; ZO

�, the government invests in violence to increase

its survival probability, but the opposition does not invest in con�ict. It isnatural to label this case government repression. Finally, when Z > ZO, theopposition mounts an insurgency, which is met with force by the incumbentgroup, and we have civil war.Our results have some striking empirical implications, when the logic of

political violence is expressed as a function of the latent variable Z: Moreprecisely, our theory predicts an ordering in Z of the three violence states:peace, repression, and civil war:12 The existing literatures on political vi-olence has studied repression and civil war as separate phenomena.13 Ourmodel provides a means of thinking about their common roots.Four empirical corollaries to Proposition 2 are worth noting in the case

where political institutions are not cohesive and/or � is far enough from oneso that political violence is a possibility, i.e. the case where Proposition 1 doesnot apply. First, higher wages, !(�1); reduce the likelihood that an economywill experience either repression or civil war since an increase in the wageincreases the opportunity cost of �ghting. Second, higher natural resourcerents, or other exogenous forms of income such as aid increase the likelihoodthat an economy will be in repression or civil war. Formally, this is because Zis higher. Third, higher expected spending on common interest public goods,due to higher � decreases the likelihood that an economy will be in repressionor civil war. This is because Z is lower when this is the case. Fourth, politicalinstitutions with more checks and balances (more minority representation)leading to a higher value of �; decrease the likelihood of observing repressionor civil war.12These are explored empircally in Besley and Persson (2010).13Davenport (2007) reviews the literature on the former and Blatman and Miguel (2009)

the literature on the latter.

14

2.4 Equilibrium Political Turnover

We can now de�ne the equilibrium rate of political turnover. Plugging in theNash equilibrium values, the political replacement rate for the period-oneincumbent group is:

� (Z; �; �) =

8>><>>: �L̂O; L̂I ; �

�Z > ZO(�; �; �)

�0; L̂I ; �

�ZO(�; �; �) � Z > ZI(�; �1; �)

(0; 0; �) ZI(�; �1; �) � Z .

(4)

The turnover rate depends on which of the three cases in Proposition 2 isrelevant, since these determine the investments in political violence. Whileit is clear that the probability of political replacement is lower in repressionthan peace, the comparison of either peace or repression with civil war isambiguous.We now draw out some comparative statics, which prove useful in our dis-

cussion of development policy and which link the rate of political replacementto Z; � and �:

Proposition 3 Suppose that Assumptions 1 and 2 hold. Then, the proba-bility of political replacement varies with fZ; �; �g as follows:

1. An increase in Z reduces the probability of political replacement whenthere is repression or civil war.

2. An increase in � increases the probability of political replacement whenthere is a civil war.

3. An increase in � reduces the probability of political replacement whenthere is repression or civil war.

The proof is in the Appendix.This is an important result when it comes to studying the impact of vio-

lence on state-capacity investments since the likelihood that the incumbentholds on to power a¤ects these decisions when � is low. The �rst part followsfrom the fact that the incumbent �ghts relatively harder than the oppositionwhen more is at stake. This is because I rises faster than O by Assumption1c. The second part of the proposition follows from the fact that it becomesmore costly for the opposition to use violence as � goes up. This reduces

15

turnover when there is a civil war, as the opposition �ghts less intensively. Itis also shifts up the threshold ZO where civil war breaks out. As for the thirdpart, the range of Z for which repression becomes wider as � increases. Thisis because, by Assumption 2, a higher value of � cuts the incumbent�s trig-ger point for violence and raises the opposition�s trigger point for violence.Within the repression and civil war regimes, a higher value of � makes theincumbent invest more in violence while the opposition invests less. Hence,it raises the probability that the incumbent group stays in power.

2.5 Investments in state capacity

To conclude the analysis of the equilibrium, we now consider investmentsin state capacities f�2; � 2g. These are determined simultaneously with theviolence decisions and maximize (2) taking LO as given.Choosing f� 2; �2g to maximize (2) yields the following Euler equations

for legal and �scal capacity:

!�(�2)[1 + (E(�2;Z; �; �; �)� 1)� 2] 0 �1L� (�2 � �1) (5)

c.s. �2 � �1 > 0

!(�2)[E(�2;Z; �; �; �)� 1] 0 �1F� (� 2 � � 1) (6)

c.s. � 2 � � 1 > 0 ,

whereE(�2;Z; �; �; �) = ��H + (1� �)E(�2j�L;Z; �; �; �)

is the expected value of period-2 public funds with

E(�2j�L;Z; �; �; �) =��L if �L � 2(1� �)2[(1� �)(1� � (Z; �; �)) + �� (Z; �; �)] otherwise .

These conditions set the marginal bene�t from investments in state capacity,which depend on the expected marginal value of public funds. To understandthe latter, observe that if �L � 2(1 � �), all spending is on the public goodregardless of the state. However, if �L < 2(1 � �), then when �s = �L,then the state spends on transfers. The expected value of those transfersis 2[(1 � �)(1 � � (Z; �; �)) + �� (Z; �; �)] which depends on the strength ofinstitutions, �, and the equilibrium rate of political survival (1�� (Z; �; �)).

16

A lower probability of political replacement increases the value of publicfunds.Equations (5) and (6) both illustrate the importance of the future value

of public funds in determining investment incentives. For �scal capacity, thisis clear as it pays o¤ in the form of an ability to raise more public funds inperiod 2. But the value of public funds matters also for legal capacity, as itcarries an indirect bene�t from investing through increased public revenues.If E(�2;Z; �; �; �) > 1, then �scal and legal capacity are complements in thesense that an increase in the stock of one kind of state capacity raises themarginal return to investing in the other kind.To pin down state capacities, we need two conditions. The �rst one is:

Cohesiveness �L � 2 (1� �) .

This requires that � be close enough to 1=2. Observe that, by Proposition 1,the cohesiveness condition also guarantees that the equilibrium is peaceful.The condition essentially guarantees that all government spending falls onpublic goods whatever the realization of �s.The second condition is:

Stability ��H + (1� �) 2 [(1� � (Z; �; �)) (1� �) + � (Z; �; �) �] � 1.

This condition says that the expected value of public spending is above unity(the value of private consumption) in a world where the cohesiveness con-dition fails and hence spending falls on transfers when �2 = �L. Whetherthis condition holds depends on the equilibrium level of political turnover.This is because transfers are more valuable to the incumbent ex ante whenhis group is more likely to retain o¢ ce, which is true when � (Z; �; �), theprobability of being replaced by the other group, is low.We now show that the model implies three types of states, when it comes

to investing in state capacities.The �rst possibility is a common interest state:

Proposition 4 If Cohesiveness holds or � is close to one, there is a common-interest state. Then,

1. there are investments in both kinds of state capacity

2. an increase in � raises both �scal-capacity and legal-capacity invest-ments, whereas changes in R; �; or � have no e¤ects on investments.

17

For this result to hold, it is su¢ cient that � is close enough to one half, or �has approached 1, so that all future marginal public revenues are allocated topublic goods. If so, the incumbent in period 1 is reassured that the state willuse public resources for common-interests, i.e., public goods regardless of whois in power in period 2. Moreover, E(�2;Z; �; �; �)= ��H +(1� �)�L > 1 sofuture public funds are valuable enough to make a positive marginal return toinvestment in �scal capacity. A higher value of � raises investments in bothaspects of that state by making future government revenue more valuableand, given that state capacities are complements raises investment in both�scal and legal capacity. Finally, by Proposition 1, common-interest statesare always peaceful, since there is no redistribution to �ght about.The second possibility is a redistributive state:

Proposition 5 If Cohesiveness fails but Stability holds, the state is redis-tributive with public revenues used to �nance transfers when �s = �L. Then,

1. there are investments in both kinds of state capacity.

2. an increase in � raises both �scal-capacity and legal-capacity invest-ments, as do (weakly) higher values of R; � or �:

In a redistributive state, the incumbent government uses available fundsto make transfers when �s = �L. It now chooses to invest in state capacity,as it is su¢ ciently likely to stay in power and to have use of that capacityas an incumbent. The stability condition implies that E(�2;Z; �; �; �) > 1so that investment in �scal capacity and legal capacity are both worthwhile.Moreover, both types of state capacity are complements.14 If the incumbent�nds itself under the risk of repression or civil war �i.e., Z is above one orboth of trigger points ZI and ZO �parameter changes that raise the intensityof repression or civil war, and hence raise the chances that the incumbentsurvives, promotes higher investments in both �scal and legal capacity. Astronger redistributive state may thus go hand in hand with more repression.Finally, we have the possibility of a weak state:

Proposition 6 If both Cohesiveness and Stability conditions fail, the stateis weak. Then, there is no incentive at all to invest in �scal capacity and the

14This complementarity is reinforced by the fact that higher state capacity of eitherkind increases Z and hence (by Proposition 3) reduces the probability of being replacedin o¢ ce.

18

level of legal-capacity investment is lower than with a common interest orredistributive state, all else equal.

The fact that the stability condition fails now implies that the mar-ginal bene�t of investing in �scal capacity is negative, E(�2;Z; �; �; �) < 1.The non-cohesiveness of political institutions and the high rate of politicalturnover means that any �scal capacity investments are likely to be used bythe other group for redistributive purposes once it takes o¢ ce. This detersinvestment in the state and we see a weak state together with high politi-cal instability associated with political violence. Legal capacity investmentis lower, because there is longer a positive bene�t from raising wages comingthrough the government budget constraint.

2.6 The State Space

Based on the results in this section, we can discuss the causes and con-sequences of state fragility. Our model shows clearly how the two mainpathologies of states, political violence and low state capacity, have commonroots. It ties these pathologies together, particularly through parameters �and �. High �; re�ecting cohesive political institutions, and high �; re�ectingstrong common interests, lead to high investments in �scal and legal capacity,as well as low violence. This is the case of a peaceful and prosperous state �the happy families in our quote from Anna Karenina in the introduction.But if � and � are low, the details of the state pathologies matter, leaving

open the possibility of low investments in �scal and legal capacity, as wellas repression or con�ict �these are the unhappy families which are unhappyin their own way. Speci�cally, parameters a¤ecting con�ict (R; �; �) becomerelevant to the outcome.The results are summarized in Table 1. Whether repression is associated

with a weak or redistributive state depends on equilibrium political stability.When �ghting is costly (high �) and the advantage of the incumbent is large(high �), we expect a redistributive state and repression to go together. Thisis the case of an e¤ective state in repression terms, which holds an incumbentin power despite weak political institutions.Civil war is generally associated with weak or redistributive regimes.

Weak states and civil wars arise under circumstances in which an insurgencyis relatively easy to organize (low �) and the government is not e¤ective inresponding to it if it happens (low �). An increase in military e¤ectiveness

19

reduces the prospect of turnover and increases incentives to invest as in aredistributive state. Another decisive factor for which particular pathologywe will observe is the amount of resource rents or foreign (cash) aid. For ex-ample, a higher degree of resource dependence (high R) raises the likelihoodthat we see civil war rather than repression.The approach taken here implies that the fundamental determinants of

fragile states are the strength of common interests, the extent of cohesiveinstitutions, the amount of resource rents, and the technologies for organizingand conducting violence. Phenomena like civil war, repression, low incomeper capita, low spending on common-interest goods, low taxation and weakenforcement of property rights are all symptoms rather than determinants.Most of the existing literature on fragile states is not derived from any

underlying theory, which explains why it tends to mix up symptoms anddeterminants. For example, one criterion that is frequently used in fragilestate indices is low income per capita. While it is true that this may increaseincentives for violence, all else equal, it is only an intermediate factor.15

An advantage of putting together a speci�c dynamic theoretical struc-ture �however simple � is that we get a clear sense of the margins wherewe may see an equilibrium response, when an outside donor intervenes infragile states. In the next section, we thus use our framework to explorethe consequences of state fragility for development policy. We will �nd thatcommon-interest and peaceful states make the task of supporting develop-ment quite easy. However, once we leave this fortunate state of a¤airs, thedetails of the pathologies matter.

3 Development Assistance

As Table 1 illustrates, a state away from the top right-hand corner, mayhave a variety of potential symptoms and these have a variety of causes.Moreover, the problems may change in response to shocks (such as hikes inresource prices or natural disasters). E¤ective development assistance has totailor the right form of intervention to circumstance and institutional context.This opens up the menu of possibilities to include the right mix of budgetary,project, military and technical assistance, and to make the right amount ofconditionality credible.

15This argument is reinforced further by introducing private capital formation as inBesley and Persson (2011, Chapter 5).

20

In our analysis, we assume that the primary objective of the internationalcommunity is the ex ante welfare of the citizens of a country to which it is pro-viding development assistance. This neglects the role of strategic objectives,which could explain the willingness of countries to donate.16 We also as-sume away coordination problems, by analyzing a single intervention, ratherthe plethora of sometimes uncoordinated action that characterizes the aidindustry. Our perspective thus supposes that a foreign government or mul-tilateral organization makes a transfer of resources to a developing country.The question is how this a¤ects the behavior of the receiving governmentand, ultimately, the welfare of the citizens.Our model suggests a number of natural margins to focus on. First, the

policy-dimension:�gs; r

Is ; r

Os

�. Does development assistance increase or de-

crease spending on public goods or the amount of redistributive transfers?Second, the state capacity dimension: (�s; � s). In which way do di¤erentforms of development assistance in�uence the incentive to build �scal or le-gal capacity? Third, the political violence dimensionn:

�LI ; LO

�. How does

development assistance a¤ects incentives of both the incumbent and opposi-tion to use violence as a means of winning or securing power?

3.1 Cash Aid

We will proceed by applying a cost-bene�t style analysis to the model. Sup-pose that an aid agency is considering spending some aid resources in aparticular country at date s. Let the shadow price of this aid in the donorcountry be �̂. Depending on circumstances and institutions, the additionalresources will raise public-goods spending, state-capacity investments, trans-fers (or, in a more general model, lower taxation). We will model cash aidas an increase in the recipient government�s budget which we denote by �R.This acts like an increase in Z in the model above.The timing of the model is exactly as in Section 3, except that in between

stages 2 and 3 the aid agency commits resources �R for future aid that willaugment the period-2 budget. In deciding whether to o¤er aid, we assumethat the aid agency can see through the subsequent equilibrium choices bygovernment thus correctly anticipating the recipient country�s responses. Ine¤ect, the aid agency is applying backwards induction to the subsequentmoves in the policy and investment game.

16See the discussion in OECD (2010).

21

We begin with the following benchmark result:

Proposition 7 In a common-interest state, cash aid is worthwhile if andonly if ��H + (1� �)�L > �̂.

The logic of this result is clear. If 2 (1� �) � �L, all future spending isdevoted to public-goods regardless of the realization of �2. There is alsono con�ict risk in this case (as showed in Proposition 1). In ex ante terms,therefore �R will be spent on public goods, with value ��H + (1� �)�L.This is compared to the cost of �̂.In this case, observe that it would not make any di¤erence whether devel-

opment assistance comes as budgetary aid or as direct support of a speci�cproject. There is a complete congruence of interest between the aid donorand the recipient government.Suppose instead that the cohesiveness condition fails, but there is no

propensity to political violence. When �2 = �L, the additional resources arespent on transfers rather than public goods. In this case, Proposition 7 ismodi�ed to:

Proposition 8 In a weak or redistributive, but peaceful, state, cash aid isworthwhile if and only if ��H + (1� �) > �̂.

In this case, the value of aid is lower than in the benchmark case of Propo-sition 7 and aid does not yield a gross return above unity � the marginalutility of consumption �as the probability of a high value of public goods�! 0:17 This result chimes with the frequently made observation, discussedin Collier and Dollar (2004), that aid impact is better when institutions arestronger.Propositions 7 and 8 allow us to re�ect upon an observation made by

Peter Bauer, which Temple (2010) has christened the Bauer Paradox. Hisview is succinctly stated in the following quote:

�A government unable to identify ... projects or collect taxes isunable to be able to use aid productively� (Bauer (1975), page400).

17In the benchmark model, granting cash aid in a common interest state makes nodi¤erence to investments in state capacity. But this is due to the assumption that utilityis linear in public goods. In the case where V (g) is increasing and concave, this propertyof the model is no longer true, and the analysis captures some aspects of the more scepticalview on aid and its impact. This is explored in Besley and Persson (2011, Chapter 6).

22

Being able to identify projects is like having high �H and/or high �.Governments that are able to collect taxes (have high �scal capacity) willlikely have more consensual institutions (Proposition 4). Hence, these arethe governments where Proposition 7 applies. But when � is low and � islow, aid is less likely to be used productively and the government is lesslikely to build �scal capacity.We now consider what happens in the case where the cohesiveness condi-

tion does not hold and where � is low enough or R high enough, so that thestate is prone to political violence. This gives two additional considerationsin the comparative statics assessing the e¤ect of cash aid on welfare. First,there is an impact on the use of political violence, and second there is an ef-fect, working through political stability, on investments in state capacities.The e¤ect of cash aid is like an increase in Z; which �as we know, from

Proposition 3 �raises the use of violence. This has two welfare e¤ects. First,it leads to more resources being allocated to violence, an activity whichis directly unproductive and has no direct welfare bene�ts. Second, whendeciding on violence, each group does not internalize this e¤ect on the welfareof the other group, leading to strategic ine¢ ciency. We summarize this as:

Proposition 9 In a weak or redistributive state, which is prone to politicalviolence, a small increase in cash aid is welfare improving if ��H+(1� �)�! (�1)

dLdZ> �̂ where

dL

dZ=

( h�1

dLI

dZ+ � dL

O

dZ

iif Z > ZO(�; �; �)

�1dLI

dZif ZO(�; �; �) � Z > ZI(�; �1; �):

With �2 = �H , we have a common-interest state and resources are spenton public goods. When �2 = �L, resources are spent on transfers. The keypoint is now that aid will have an impact on equilibrium violence re�ectedin the third term deducted from the value of any public goods that are beingcreated. Compared to Proposition 8, therefore, it is less likely that cash aidis welfare increasing due to the additional welfare cost of violence. Indeed,cash aid could actually lower ex ante welfare. This is more likely when � islow.The enhanced violence a¤ects political stability in line with the results in

Proposition 3.

Proposition 10 In a weak or redistributive state, which is prone to politicalviolence, cash aid can increase political stability and may increase investmentin �scal and legal capacity.

23

This e¤ect comes through the fact that foreign aid (by Proposition 3) allowsincumbents to entrench themselves and cement their control on power, wheninstitutions are non-cohesive. The most clear-cut example is the case of arepressive regime. Once aid is given, the incentive to hang on to power isenhanced. As we have seen, in Proposition 5, greater repression inducesmore political stability, which results in more state-capacity investment allelse equal. But the bene�ts are allocated in part to increased military forceand accrue disproportionately to the incumbent group.

3.2 Conditionality

The assumption in the previous subsection is that government cannot con-tract directly over the policy and investment decisions when it grants aid.Conditionality should be thought of as a contracting problem where the donorgovernment speci�es an array of observable and veri�able decisions by therecipient government in exchange for �R. As with any interesting contract-ing problem, the real issue is what can reasonably supposed to be observedand enforced. The latter is a particular issue, given that there is nothingequivalent to an international court which can enforce agreements. Indeed,this is often thought of as a major obstacle to e¤ective conditionality.18

It is interesting to see how the earlier results might be a¤ected by (en-forceable) conditionality. Propositions 8 and 9 highlight the possibility thatconditionality to make sure that aid is spent on public goods could be valu-able. In the case of Proposition 9, it would ensure that fewer resources arespent on violence. However, for this latter e¤ect, it would have to be the casethat conditionality is binding on both the incumbent and the opposition ifthe latter wins o¢ ce in the struggle for power.But these prospective bene�ts require that conditionality could be credi-

bly enforced. An interesting agenda for future research would be to combineaspects of our framework with explicit modeling of mechanisms that mighthelp achieve credible enforcement, as in Svensson (2003). External playerspotentially have several prospective instruments, whereby they can give afuture prize in return for present and continued good behavior. This couldinclude membership in or association with desirable clubs like EU (some-thing used by EU to in�uence political and economic reform in Turkey and

18See Svensson (2000, 2003) for early analyses of the credibility problems wih condition-ality.

24

earlier in Eastern Europe). Another possibility might involve debt conces-sions (or free-trade agreements) like the ones currently negotiated by EUwith Pakistan and with some African countries, and by the US with severalLatin-American countries.

3.3 Non-Cash Development Assistance

In this section, we consider the possibility of development assistance in otherforms than cash aid, as captured by in�uence on other parameters of themodel.

Technical Assistance Technical assistance refers to the transfers of skillsand knowledge that can be useful in improving how government works. Es-timates from the OECD (2010b) suggest at least one sixth of direct o¢ cialaid in 2008 is given in this form. Some people refer to technical assistanceas phantom aid, since it is often dispersed via international consultants whoreside in the donor country. Evaluating the returns to technical assistance isnotoriously di¢ cult, and is likely to be speci�c to the context and the na-ture of the intervention in question. Technical assistance can come in manyforms. Our model suggests a focus on two of these: (i) e¤orts to increasethe bene�ts or reduce the costs of providing public goods, and (ii) e¤orts tolower the cost of investing in state capacity. We will discuss these two caseshere and demonstrate how their impact can be thought of in our analyticalframework.Consider �rst technical assistance that helps to identify good projects.

This can be represented in the model as an attempt to raise � or �. Animportant line of development research in recent years has used randomizedcontrol trials (RCTs) to identify the value of public interventions.19 These canbe thought of as trying to �nd ways of better allocating resources to publicgoods by identifying high bene�t interventions. Such interventions also raisethe investments in state capacity (by Propositions 4 and 5). For given �,it also makes it less likely that there will be con�ict (by Proposition 2) andincreases the probability of creating a common-interest state. We record thisas:

Proposition 11 Technical assistance that raises �H or �, increases welfare

19See Du�o et al (2007) for a discussion of these methods.

25

and investment in state capacity. It also reduces the likelihood of politicalviolence.

Technical assistance can also be aimed towards increasing �L: In our frame-work this could even lead to the creation of a common-interest state with itsvirtuous consequences. In the case describe here, the cost of the interventionshould be weighed against the increased public goods that will be provided(since �scal and legal capacity increase) and the value of the extra publicgoods that are generated by these investments.Our model shows the same complementarity between the value of aid and

cohesive institutions that we saw in the case of cash aid. Technical assistanceis more powerful in countries where public resources are more likely to beallocated for the common good. So, the return to RCTs will be lower in weakor redistributive states if they are intended to inform government about thevalue of good policy.While these observations are useful, they sweep a host of important prob-

lems under the rug. First, there is a scaling-up issue. Can the conditions insmall controlled trials really be replicated when they are implemented in alarge program by the government? This question is particularly acute sincemany RCTs are implemented directly by NGOs, with limited contact withthe recipient government. Second, in assuming that the government budgetwill be spent either on public goods or transfers, depending on the value of �;we have not considered any issues of corruption or predation run by a smallelite (see Section 4).Another type of technical assistance would be to reduce investment costs �

improving state capabilities. These interventions are common in developmentand can be represented in the model as shifts in the functions F (�) and L (�).In this case, we have:

Proposition 12 Technical assistance that reduces the cost of investing instate capacity, i.e. reduces F (�) and L (�), increases welfare and investmentin state capacity. This tends to increase the likelihood of political violence, allelse equal.

The result on state capacity follows from the Euler equations (5) and (6).Our framework makes sense of these type of interventions. Examples in-clude giving advice on tax collection, or the creation of specialized courtsto expedite the resolution of business disputes. It could even mean advice

26

about fundamental changes in the nature of the legal code. The (perhapssurprising) e¤ect on political violence comes from the fact that Z goes upwhen state-capacity investments increase which, all else equal, may lead toan increase in political violence

Military assistance We now consider the role of military assistance withinthe con�nes of our simple model. One aspect of such assistance may be togive advice on military technology/strategy, which can be modeled as changesin �. This could be training or provision of weapon systems and intelligence.In principle, this could be o¤ered either to the government or the opposition.External governments could also directly intervene by providing manpowerto either the government or rebels. We focus on the case, where support iso¤ered to the incumbent government.Our core result is:

Proposition 13 Military assistance that increases �; augmenting the mili-tary capacity of the incumbent government, increases the parameter range inwhich there is repression. This increases political stability and investment in�scal and legal capacity.

If institutions are weak, the higher political stability due to higher repressionsraise state capacity investments (by Propositions 3 and 5). But this comesat the price of increasing the entrenchment of the incumbent. In e¤ect, thiscan create a rentier state, where the opposition group is frozen out of power.Military intervention to help any incumbent will therefore tend to increasethe incentives for the incumbent to invest. But it is not a substitute formore consensual institutions (higher �). This story seems relevant, perhapsespecially for the cold war, but also for modern-day fragile states with ongoingor latent con�ict.

Post-con�ict assistance Finally, we consider attempts by external actorsto assist in the process of promoting peace in post-con�ict situations. Ourmodel allows us to represent this in a very stylized fashion.Peace-keeping or disarming the rebels can be thought of as raising �:

This reduces the parameter range in which there is con�ict and increasespolitical stability (by Proposition 3). Many post-con�ict settlements can

27

also be thought of as e¤orts to raise � (diminishing the gain to the winner).20

This will reduce the risk of violence (by Propositions 1 and 2). However, thelatter e¤ect requires that the interventions are expected and hence credibleex ante. A recent example of such a mechanism was attempted in Haitiafter the recent earthquake, where the high in�ux of aid has been disbursedoutside of the government structures with former U.S. President Bill Clintonplaying a key role. We summarize this in:

Proposition 14 Post-con�ict assistance that raises � or � will lead togreater investments in state capacities and reduce the parameter range inwhich there is violence.

Of course, post-con�ict reconstruction may have a wider remit part of whichcould involve direct e¤orts to increase � and �. However, post-con�ict assis-tance generally comes with a good deal of cash aid, making reforms to raise� and lower � doubly important.A related question is if we can think about a window of opportunity in

weak or collapsed states. Can external players assist in designing a politi-cal mechanism and/or institutions capable of changing the internal politicalequilibria in countries, such as Iraq, Afghanistan, and Somalia? Historicalexperience cautions us that a simple export or duplication of existing in-stitutions need not work in many cases. Thus, almost all Latin Americanstates modeled their constitutions on the US one in the 19th century, butwith very di¤erent results. The commonwealth countries of India and Pak-istan started o¤ with the same constitutional framework, provided by the1935 Government of India Act, but with very di¤erent outcomes.Changing political mechanisms and institutions would be like changing �

and thereby potentially altering the type of state. To discuss this possibility,we need to think carefully about the domestic incentives for undertakingendogenous political reform. Such an extension of our analysis is brie�ydiscussed in the next section.20It may also be the case that there are measures to increase � which �t better under

technical assistance � e.g., to improve the monitoring of public resource use making itmore transparent which group is bene�tting from transfers.

28

4 Extensions

The structure we have proposed in this paper is simple and stylized and theanalysis can be extended in many directions. We now brie�y sketch �vepossibilities.

Donor preferences for distribution In Propositions 7 and 8, the ex antevalue of a dollar of transfers is always greater than one; the worst that canhappen is that aid ends up as higher private consumption. The donor countrycould however, care about the distribution of such transfers. This will tendto reduce the value of transfers, to the extent that the donor is egalitarianand � < 1=2. To illustrate this point, suppose that the donor has a maximinpreference over distribution between the two groups. Then cash aid is worthonly 2� (< 1) when �s = �L in the non-cohesive case of Proposition 8. Thenaid is worthwhile only if ��H + (1� �) 2� > �̂ and the left-hand side couldnow be smaller than one.

Quasi-linear preferences We have worked with the stylized case whereutility is linear in public goods. Suppose instead that preferences are �sV (gs)+xJs where V (�) is increasing and concave �as we mentioned as a possibilityin Footnote 16 above. As shown in Besley and Persson (2011, Chapter 6)budgetary aid now directly depresses the demand for �scal capacity since themarginal value of public goods is lower (for any given level of �scal capac-ity). Aid will also have a smaller e¤ect on public goods to the extent that itcrowds out domestically generated �scal capacity. Both of these e¤ects lowerthe impact of aid on public-good provision.

Predatory states Our model does have redistribution across groups, butassumes that the government faithfully represents the interests of the entireincumbent group. But a feature of many countries in weak states is thatpolitical elites are able to capture the state and to bene�t disproportion-ately from state transfers. Such conditions are often referred to as predatorystates.21 A very simple way to think about a predatory state in our frame-work is to suppose that only a fraction (1� {) of the resources available for21This is often more than just elite control but also refers to the kind of ine¢ cient

predatory practices that elites use to extract resources as discussed in Besley and Persson(2011, Chapters 3, 5 and 7).

29

transfers go to broad based transfers across the groups to which the cohe-siveness parameter, �; applies. The remaining fraction, {; is captured by anincumbent elite of size eI < 1. Then the payo¤to a member of the incumbentelite payo¤ is:

�sgs + (1� � s)! (�s) +h(1� {) 2 (1� �) + {

eI

i[R + � s! (�s)� gs �ms] :

If { > 0, this change in the model will a¤ect both policy and investmentincentives. For public goods to be provided, we now require that

�s � (1� {) 2 (1� �) + {=eI

as the equivalent of the cohesiveness condition. Since eI < 1, this is morestringent than the cohesiveness condition stated above and this conditionmay fail even if �s = �H and � = 1=2. Thus, it is more likely that R goesinto transfers than in the baseline model. Moreover, these transfers couldmainly be funneled to elites, depending on the value of {. Both of thesee¤ects make the prize from capturing the state greater, and this tends tointensify the incentives for political violence if we assume that violence isindeed organized by elites.

Endogenous political institutions We have taken � to be exogenous.But one critique of aid is that it leads to a deterioration in institutionalquality. The model can be extended to capture this by supposing that theincumbent can also make decisions in period one that a¤ects �2. As Besleyand Persson (2011, Chapter 7) discuss in detail, there will be a tendencytowards picking a lower �2 when is low. To the extent that an increasein R reduces , as we predict it would in Proposition 3, larger �ows of cashaid will lead to worse institutions. The key to cohesive institutions thenlies in understanding which factors might limit the ability of states to lower� and/or ; rather than in these parameters per se. According to the logicof our model, this is an important area where more work is needed, giventhe importance of cohesive institutions in reducing political violence andpromoting state-capacity investments.

Private accumulation The only engine of growth in our model is higherlevels of legal capacity. This is, of course, an abstraction to focus on a speci�cmechanism working through investments in the state and their consequences.

30

The model can also be extended to include �standard�dynamic incentives toincrease income per capita such as the accumulation of capital. The analysisin Besley and Persson (2011, Chapters 3 and 5) suggests that the pathologiesof the state that we have identi�ed here will generally spill over into weakerincentives for individuals to invest, further compounding the consequences ofstate fragility and reducing incomes. This will amplify the link between lowincomes and state fragility.

5 Concluding Remarks

We have presented a framework that makes sense of current policy debatesabout impediments to development in fragile states. While attempts to de�nestate fragility precisely are open to interpretation, they are crucial in remind-ing us that political violence and ine¤ective states are commonplace in manylow-income countries. Our analysis shows how such situations can be under-stood and highlights some features of the economic environment which mayperpetuate the problem. Table 1 summarizes our insights and underlines thecommon roots of state pathologies. In broad terms, the deepest root is theabsence of common interests reinforced by non-cohesive institutions. Havingunderstood this, the question of what determines the cohesiveness of institu-tions and whether this can be changed becomes of �rst-order importance.It is important to acknowledge that this paper is a purely theoretical

exercise. In general, we know precious little about the empirical importanceof the channels identi�ed by our theory, and even less about their importancein the speci�c states plagued by the pathologies highlighted in the paper.Better knowledge can only be acquired with appropriate empirical analysison a case-by-case basis.OurAnna Karenina Principle underlines the importance of heterogeneity.

This heterogeneity may, to some degree, reconcile the di¤erent positions inthe debate on development policy which we discussed in the introduction.Our model suggests, along with aid optimists, that there is the possibilityof advantageous development assistance. But identifying the form that thiswill take requires a great deal of knowledge about country circumstancesand institutions. One could thus be an aid pessimist or an aid optimist,depending on the form of aid and an assessment of the ability of aid agenciesto understand its impact in speci�c contexts.

31

References

[1] Acemoglu, Daron (2005), �Politics and Economics in Weak and StrongStates�, Journal of Monetary Economics 52, 1199-1226.

[2] Alesina, Alberto, and David Dollar, (2000), �Who Gives Foreign Aid toWhom and Why?�Journal of Economic Growth 5, 33-63.

[3] Bates, Robert H.:(2008), "The Logic of State Failure: Learning fromLate-Century Africa", Con�ict Management and Peace Science 25, 297-314.

[4] Bauer, Peter (1972). Dissent on Development, Cambridge: Harvard Uni-versity Press.

[5] Bauer, Peter (1975), �N.H. Stern on Substance and Method in Devel-opment Economics,�Journal of Development Economics 2, 387-405.

[6] Besley, Timothy and Torsten Persson (2009), �The Origins of State Ca-pacity: Property Rights, Taxation and Politics�, American EconomicReview 99, 1218-1244.

[7] Besley, Timothy and Torsten Persson (2010), �The Logic of PoliticalViolence�, forthcoming in the Quarterly Journal of Economics.

[8] Besley, Timothy and Torsten Persson (2011), Pillars of Prosperity, man-uscript based on 2010 Yrjö Jahnsson Lectures, Princeton UniversityPress, forthcoming.

[9] Blattman, Christopher and Edward Miguel (2009), �Civil War,�Journalof Economic Literature 48, 3-57.

[10] Caselli, Francesco (2006), "Power Struggles and the Natural ResourceCurse", mimeo, LSE.

[11] Chenery, Hollis B. and Alan M. Strout (1966), �Foreign Assistance andEconomic Development,�American Economic Review 56, 679-733.

[12] Chauvet, Lisa and Paul Collier (2004), �Development E¤ectiveness inFragile States: Spillovers and Turnarounds.�Oxford, UK: Oxford Uni-versity, Center for the Study of African Economies, Department of Eco-nomics, January.

32

[13] Chauvet, Lisa and Paul Collier, Paul, (2006), �Helping Hand? Aid toFailing States�DIAL - Document de Travail DT/2006-14.

[14] Collier, Paul (2007), The Bottom Billion, Oxford: Oxford UniversityPress.

[15] Collier, Paul and David Dollar (2004), �Development E¤ectiveness:What Have We Learnt?,�Economic Journal 114, F244-F271.

[16] Collier, Paul, V. L. Elliott, Håvard Hegre, Anke Hoe­ er, MarthaReynol-Querol, and Nicholas Sambanis (2003), Breaking the Con�ictTrap: Civil War and Development Policy, New York: Oxford Univer-sity Press, for the World Bank.

[17] Davenport, Christian (2007), �State Repression and Political Order,�Annual Review of Political Science 10, 1-23.

[18] Djankov, Simeon, Jose G. Montalvo and Marta Reynal-Querol (2008),�The Curse of Aid,�Journal of Economic Growth 13, 1835-1865.

[19] Du�o, Esther, Rachel Glennerster and Michael Kremer, (2007), �UsingRandomization in Development Economics Research: A Toolkit,�CEPRDiscussion Paper, No 6059.

[20] Easterly, William (2006), The White Man�s Burden: Why the West�sE¤orts to Aid the Rest Have Done So Much Ill and So Little Good,Penguin Books.

[21] Easterly, William, (2003) �Can Foreign Aid Buy Growth?,�Journal ofEconomic Perspectives, 17(3), 23-48.

[22] Gupta, Sanjeev, Catherine Pattillo and Smita Wagh [2006], �Are DonorCountries Giving More or Less Aid?�, Review of Development Eco-nomics 10, 535-552.

[23] McGillivray, Mark (2006), �Aid Allocation and Fragile States�, WIDERDiscussion Paper 2006/01, Helsinki.

[24] OECD, (2010a), Do No Harm: International Support for State Building,Paris: OECD publications.

33

[25] OECD (2010b), Development Cooperation Report, Statistical Annex,OECD, Paris.

[26] Rice, Susan and Stewart Patrick (2008), Index of State Weakness in theDeveloping World, Washington, DC: The Brookings Institution.

[27] Riddell, Roger (2007), Does Foreign Aid Really Work?, Oxford: OxfordUniversity Press.

[28] Sachs, Je¤rey (2005), The End of Poverty: Economic Possibilities forOur Time, Penguin Books.

[29] Svensson, Jakob (2000), �When is Foreign Aid Policy Credible? AidDependence and Conditionality�, Journal of Development Economics61, 61-84.

[30] Svensson, Jakob (2003), �Why Conditional Aid Doesn�t Work andWhatCan Be Done About It?�, Journal of Development Economics 70, 381-402

[31] Temple, Jonathan (2010). �Aid and Conditionality.� in Rodrik, Daniand Mark Rosenzweig (eds.), Handbook of Development Economics vol.5, 4415-4523.

[32] U.S. Agency for International Development, (2005), Fragile States Strat-egy, available at www.usaid.gov

[33] Weingast, Barry (2005), �The Constitutional Dilemma of Economic Lib-erty�, Journal of Economic Perspectives 19, 89-108.

34

6 Proof of Propositions

Proof of Proposition 2. The complementary slackness conditions for theproblems faced by LI and LO assuming that LI > 0 and LO > 0 are:

� I�LO; LI ; �

�xZ � �1 7 0

c.s. �LI � LI � 0.

and

O�LO; LI ; �

�xZ � � 7 0

c.s. �LO � LO � 0:

where x = (1� 2�)2 (1� �) 2 [0; 2].First, we show that, at any interior solution, resources devoted to �ghting

by both groups is increasing in Z. To see this, observe that di¤erentiatingand using the �rst-order conditions when they hold with equality yields:�

� IIxZ � IOxZ IOxZ OOxZ

� �dLI

dLO

�=

IxdZ� OxdZ

. (7)

De�ne =�� II OO + ( IO)

2�x2Z2 > 0. Solving (7) using Cramer�s ruleyields:

dLI

dZ=x2Z [ I OO � O IO]

> 0

anddLO

dZ=x2Z [ II O � I IO]

> 0 .

where we have used both parts of Assumption 1d.We now derive two trigger points for violence. De�ne L̂ (Z) from

� I�0; L̂ (Z) ; �

�xZ � �1 7 0

c.s. �LI � L̂ (Z) � 0 .

It is simple to check that this is an increasing function of Z under Assumption1b. Clearly with LO = 0, LI = L̂ (Z). We can de�ne ZI(�; �) from L̂ (Z) = 0,i.e.,

ZI(�; �) =��1

I (0; 0; �)x:

35

Next, de�ne ZO(�; �) implicitly from

O

�0; L̂(ZO(�; �)

�xZO(�; �) = � :

The expression for dLO

dZimplies that for Z � ZO, we must have LO > 0.

As the next step, we prove that ZO(�; �) > ZI(�; �). Suppose not, then

O (0; 0; �)xZO(�; �) = � :

If so,

ZO(�; �) =�

O (0; 0; �)x� ZI (�; �) = ��1

I (0; 0; �)x,

or� I (0; 0; �) O (0; 0; �)

<�1�� �H

�,

which contradicts Assumption 1c for all values of �.Finally, it is easy to see from the explicit de�nition that ZI(�; �) is an in-

creasing function. Using the implicit de�nition of ZO(�; �); and the fact thatL̂�ZO(�; �)

�is (weakly) increasing, it follows that this function is increasing

as well. This concludes the proof of the proposition.Proof of Proposition 3. Part 1 is proved as follows. First, suppose thatthere is repression then the result follows since Assumption 1 implies thatthe incumbents payo¤ is concave in LI and Z increases the marginal bene�tto �ghting. Now suppose that there is civil war. Di¤erentiating (4) withrespect to Z yields:

�Z (Z; �; �) = IdLI

dZ+ O

dLO

dZ

=

�( I)

2 OO + ( O)2 II � 2 I O IO

��� II OO + ( IO)

2�Z < 0

where we have used the comparative static result in (7) and Assumption 1c.To prove part 2, observe that under repression, there is no e¤ect on L̂I

from changing � since LO = 0. Now observe that with civil war, then as �increases, the e¤ect on violence is given by:�

� IIxZ � IOxZ IOxZ OOxZ

� �dLI

dLO

�=

0d�

. (8)

36

Di¤erentiating (4) with respect to � and using Cramer�s rule in 8), we have:

��(Z; �; �) = IdLI

d�+ O

dLO

d�

=[� II O + I IO]�

� II OO + ( IO)2� 2 (1� 2�)Z2 < 0

by Assumption 1c.Now we turn to part 3. First, suppose that there is repression then

the result follows since Assumption 1 implies that the incumbents payo¤ isconcave in LI and � increases the marginal bene�t to �ghting. Now supposethat there is civil war. Then as � increases, the e¤ect on violence is given by:�

� IIxZ � IOxZ IOxZ OOxZ

� �dLI

dLO

�=

I��LO; LI ; �

�xZd�

� O��LO; LI ; �

�xZd�

. (9)

where, as above, x = (1 � 2�)2. Di¤erentiating (4) with respect to � andusing Cramer�s rule in (9), we have:

��(Z; �; �) = IdLI

d�+ O

dLO

d�

= O� [ II O � I IO] + I� [ OO I � IO O]�

� II OO + ( IO)2�Z < 0

using Assumption 1c and Assumption 2. This proves the result.

37

Table 1The state space

Weak Redistributive Common interest

Peace low �; �; Rhigh �low �; R

high �, �

Repressionlow �; �; �; Rhigh �

low �; �;Rhigh �; �

n/a

Civil warlow �; �; �high �;R

low �; �; �high �; R

n/a

38